[liangqianxing]

Problem

With default cycling, mixing an empty source with range(4) yields [], [0] or [0, 1] instead of all four records. The second StopIteration after resetting the empty source escapes the sampler. Weighted CYCLE_FOREVER is affected too.

Changes

Catch the failed restart and continue with other sources. Round-robin advances its source index; weighted sampling excludes failed restarts for the current next() call and samples directly among the remaining weights. In cycling modes, this also bounds attempts when every source is empty, including highly skewed weights such as 1e20:1.

The weighted fallback uses a lazy, checkpointed auxiliary RNG. Ordinary no-empty sampling keeps its existing batched random sequence and checkpoint fields; old checkpoints remain loadable. Exclusions are local to one call, so a source whose contents change on reset can be retried later.

Add public Loader regressions for empty-source order, finite/forever cycling, all-empty inputs, extreme weights, and serialized repeated checkpoint restore with multiple nonempty sources.

Validation

• New regressions against unmodified main: 14 failed, 4 passed.
• Final two sampler test modules: 193 passed.
• Repository pre-commit hooks for all four changed files pass.
• Independent checks compared 147 ordinary sampling sequences with main, restored 24 old and 28 new checkpoints, and exercised 7 dynamic-source cases; all passed. Conditional sampling and extreme-weight termination were checked separately.

The repository mypy command was run; its unmodified baseline has the same quoted python_version configuration warning and generated torchdata.version typing error. No new typing error was observed.

AI assistance was used to investigate, implement and test this change.

[meta-cla]

Hi @liangqianxing!

Thank you for your pull request and welcome to our community.

Action Required

In order to merge any pull request (code, docs, etc.), we require contributors to sign our Contributor License Agreement, and we don't seem to have one on file for you.

Process

In order for us to review and merge your suggested changes, please sign at https://code.facebook.com/cla. If you are contributing on behalf of someone else (eg your employer), the individual CLA may not be sufficient and your employer may need to sign the corporate CLA.

Once the CLA is signed, our tooling will perform checks and validations. Afterwards, the pull request will be tagged with CLA signed. The tagging process may take up to 1 hour after signing. Please give it that time before contacting us about it.

If you have received this in error or have any questions, please contact us at cla@meta.com. Thanks!

[meta-cla]

Thank you for signing our Contributor License Agreement. We can now accept your code for this (and any) Meta Open Source project. Thanks!